Efficient Sorting in a Dynamic Adverse Selection Model I. Hendel, A

Efficient Sorting in a Dynamic
Adverse Selection Model
I. Hendel, A. Lizzeri, and M. Siniscalchi
Review of Economic Studies (2005), 72, 467-497.
Presented by Tomek Piskorski
Motivation and Objective of the Paper
Motivation and Objective of the Paper
• Since Akerlof’s (1970) seminal paper, adverse selection
has been recognized to be a potential source of
inefficiency in durable-goods markets.
Motivation and Objective of the Paper
• Since Akerlof’s (1970) seminal paper, adverse selection
has been recognized to be a potential source of
inefficiency in durable-goods markets.
• Recent literature has pointed out that the extent of
inefficiency in the classic adverse selection model
depends on restrictions of trading opportunities.
Motivation and Objective of the Paper
• Since Akerlof’s (1970) seminal paper, adverse selection
has been recognized to be a potential source of
inefficiency in durable-goods markets.
• Recent literature has pointed out that the extent of
inefficiency in the classic adverse selection model
depends on restrictions of trading opportunities.
• Hendel and Lizzeri (1999, 2002) and Johnson and
Waldman (2003) departs from the exogenous ownership
assumption. They find that some inefficiency remain.
Motivation and Objective of the Paper
• Since Akerlof’s (1970) seminal paper, adverse selection
has been recognized to be a potential source of
inefficiency in durable-goods markets.
• Recent literature has pointed out that the extent of
inefficiency in the classic adverse selection model
depends on restrictions of trading opportunities.
• Hendel and Lizzeri (1999, 2002) and Johnson and
Waldman (2003) departs from the exogenous ownership
assumption. They find that some inefficiency remain.
• Janssen and Roy (2001, 2002) deals with the restricted
secondary markets assumption. Some inefficiency remain.
Motivation and Objective of the Paper
• Since Akerlof’s (1970) seminal paper, adverse selection
has been recognized to be a potential source of
inefficiency in durable-goods markets.
• Recent literature has pointed out that the extent of
inefficiency in the classic adverse selection model
depends on restrictions of trading opportunities.
• Hendel and Lizzeri (1999, 2002) and Johnson and
Waldman (2003) departs from the exogenous ownership
assumption. They find that some inefficiency remain.
• Janssen and Roy (2001, 2002) deals with the restricted
secondary markets assumption. Some inefficiency remain.
• Can we achieve efficiency in the adverse selection model
where both restrictions of trading opportunities are
removed at the same time?
Set-up
Set-up
Time: Discrete, Infinite Horizon (period lasts Δ ∈ 0, 1 units).
Set-up
Time: Discrete, Infinite Horizon (period lasts Δ ∈ 0, 1 units).
There is a total mass Y  1 of cars.
Set-up
Time: Discrete, Infinite Horizon (period lasts Δ ∈ 0, 1 units).
There is a total mass Y  1 of cars.
At any time, the quality of a car may take up one of finitely
many values, denoted q 0  q 1 . . .  q N ≥ 0.
Set-up
Time: Discrete, Infinite Horizon (period lasts Δ ∈ 0, 1 units).
There is a total mass Y  1 of cars.
At any time, the quality of a car may take up one of finitely
many values, denoted q 0  q 1 . . .  q N ≥ 0.
For n  0, . . . N, a newly produced car has quality q n with
N
probability  n  0, where ∑ n0  n  1.
Set-up
Time: Discrete, Infinite Horizon (period lasts Δ ∈ 0, 1 units).
There is a total mass Y  1 of cars.
At any time, the quality of a car may take up one of finitely
many values, denoted q 0  q 1 . . .  q N ≥ 0.
For n  0, . . . N, a newly produced car has quality q n with
N
probability  n  0, where ∑ n0  n  1.
For n  0, . . . , N and m  n  1, . . . , N  1, a car of quality q n
depreciates to q m (if n  m ≤ N) or dies (if m  N  1) with
probability  n,m in every time period.
Set-up
Time: Discrete, Infinite Horizon (period lasts Δ ∈ 0, 1 units).
There is a total mass Y  1 of cars.
At any time, the quality of a car may take up one of finitely
many values, denoted q 0  q 1 . . .  q N ≥ 0.
For n  0, . . . N, a newly produced car has quality q n with
N
probability  n  0, where ∑ n0  n  1.
For n  0, . . . , N and m  n  1, . . . , N  1, a car of quality q n
depreciates to q m (if n  m ≤ N) or dies (if m  N  1) with
probability  n,m in every time period.
The simple depreciation case corresponds to  0  1 and
 n,m  0 for m  n  1.
Set-up
Set-up
There is a unit mass of infinitely lived consumers.
Set-up
There is a unit mass of infinitely lived consumers.
Consumers differ in their private valuation for quality,
a “type”  ∈ ,  is distributed according to the c.d.f. F.
Set-up
There is a unit mass of infinitely lived consumers.
Consumers differ in their private valuation for quality,
a “type”  ∈ ,  is distributed according to the c.d.f. F.
Each period consumers receive an endowment e of ‘money’.
Set-up
There is a unit mass of infinitely lived consumers.
Consumers differ in their private valuation for quality,
a “type”  ∈ ,  is distributed according to the c.d.f. F.
Each period consumers receive an endowment e of ‘money’.
The utility of a type- consumer who, at each time t, drives
a quality-qt car and pays (lump-sum) Pk at time t k is
Set-up
There is a unit mass of infinitely lived consumers.
Consumers differ in their private valuation for quality,
a “type”  ∈ ,  is distributed according to the c.d.f. F.
Each period consumers receive an endowment e of ‘money’.
The utility of a type- consumer who, at each time t, drives
a quality-qt car and pays (lump-sum) Pk at time t k is
Set-up
There is a unit mass of infinitely lived consumers.
Consumers differ in their private valuation for quality,
a “type”  ∈ ,  is distributed according to the c.d.f. F.
Each period consumers receive an endowment e of ‘money’.
The utility of a type- consumer who, at each time t, drives
a quality-qt car and pays (lump-sum) Pk at time t k is
where  is the instantaneous discount rate.
Set-up
There is a unit mass of infinitely lived consumers.
Consumers differ in their private valuation for quality,
a “type”  ∈ ,  is distributed according to the c.d.f. F.
Each period consumers receive an endowment e of ‘money’.
The utility of a type- consumer who, at each time t, drives
a quality-qt car and pays (lump-sum) Pk at time t k is
where  is the instantaneous discount rate.
We assume that e is finite and large enough that consumers
can potentially afford any quality they wish.
Steady State Efficiency (Efficient Sorting)
Steady State Efficiency (Efficient Sorting)
Let v ∗n denote the steady-state mass of cars of quality q n .
Steady State Efficiency (Efficient Sorting)
Let v ∗n denote the steady-state mass of cars of quality q n .
For n  0, . . . , N, the steady-state mass v ∗n must satisfy the
following system of equations:
Steady State Efficiency (Efficient Sorting)
Let v ∗n denote the steady-state mass of cars of quality q n .
For n  0, . . . , N, the steady-state mass v ∗n must satisfy the
following system of equations:
Steady State Efficiency (Efficient Sorting)
Let v ∗n denote the steady-state mass of cars of quality q n .
For n  0, . . . , N, the steady-state mass v ∗n must satisfy the
following system of equations:
Steady State Efficiency (Efficient Sorting)
Let v ∗n denote the steady-state mass of cars of quality q n .
For n  0, . . . , N, the steady-state mass v ∗n must satisfy the
following system of equations:
The ex-post efficient steady-state allocation of cars to
consumers can be described via cutoff types  ∗0 , . . . ,  ∗N s.t.
Steady State Efficiency (Efficient Sorting)
Let v ∗n denote the steady-state mass of cars of quality q n .
For n  0, . . . , N, the steady-state mass v ∗n must satisfy the
following system of equations:
The ex-post efficient steady-state allocation of cars to
consumers can be described via cutoff types  ∗0 , . . . ,  ∗N s.t.
- types  ∈  ∗0 ,  hold a quality-q 0 , and their mass equals v ∗0 ,
Steady State Efficiency (Efficient Sorting)
Let v ∗n denote the steady-state mass of cars of quality q n .
For n  0, . . . , N, the steady-state mass v ∗n must satisfy the
following system of equations:
The ex-post efficient steady-state allocation of cars to
consumers can be described via cutoff types  ∗0 , . . . ,  ∗N s.t.
- types  ∈  ∗0 ,  hold a quality-q 0 , and their mass equals v ∗0 ,
- types  ∈  ∗n ,  ∗n−1  hold a quality-q n and their mass equals v ∗n .
Simple Depreciation Model:
Observable Quality Benchmark
Simple Depreciation Model:
Observable Quality Benchmark
Under rental implementation:
Simple Depreciation Model:
Observable Quality Benchmark
Under rental implementation:
Consumer who rents a car of quality q n pays an
instantaneous rental price r n throughout the period.
Simple Depreciation Model:
Observable Quality Benchmark
Under rental implementation:
Consumer who rents a car of quality q n pays an
instantaneous rental price r n throughout the period.
The rental prices that sustain the efficient allocation satisfy:
Simple Depreciation Model:
Observable Quality Benchmark
Under rental implementation:
Consumer who rents a car of quality q n pays an
instantaneous rental price r n throughout the period.
The rental prices that sustain the efficient allocation satisfy:
Simple Depreciation Model:
Observable Quality Benchmark
Under rental implementation:
Consumer who rents a car of quality q n pays an
instantaneous rental price r n throughout the period.
The rental prices that sustain the efficient allocation satisfy:
where by convention r ∗N1  q N1  0.
Simple Depreciation Model:
Observable Quality Benchmark
Under rental implementation:
Consumer who rents a car of quality q n pays an
instantaneous rental price r n throughout the period.
The rental prices that sustain the efficient allocation satisfy:
where by convention r ∗N1  q N1  0.
Under selling implementation:
Simple Depreciation Model:
Observable Quality Benchmark
Under rental implementation:
Consumer who rents a car of quality q n pays an
instantaneous rental price r n throughout the period.
The rental prices that sustain the efficient allocation satisfy:
where by convention r ∗N1  q N1  0.
Under selling implementation:
The prices p n that sustain the efficient allocation are
defined by the expected present value of rental prices.
Simple Depreciation Model:
Observable Quality Benchmark
Simple Depreciation Model:
Observable Quality Benchmark
Under both implementaions the quality of any given unit of
the good offered on the market can be exactly inferred from
its vintage: the number of times the unit has changed hands.
Simple Depreciation Model:
Observable Quality Benchmark
Under both implementaions the quality of any given unit of
the good offered on the market can be exactly inferred from
its vintage: the number of times the unit has changed hands.
A unit of vintage n is of quality q n .
Simple Depreciation Model:
Observable Quality Benchmark
Under both implementaions the quality of any given unit of
the good offered on the market can be exactly inferred from
its vintage: the number of times the unit has changed hands.
A unit of vintage n is of quality q n .
The equilibrium strategies can be formulated as follows:
∗
 rent or buy vintage-n cars,
Consumer types  ∈  ∗n ,  n−1
and keep the same unit until it depreciates.
Simple Depreciation Model with Unobservable Quality
Simple Depreciation Model with Unobservable Quality
Consumers cannot observe the quality of a car without using it.
Simple Depreciation Model with Unobservable Quality
Consumers cannot observe the quality of a car without using it.
Theorem 1 :
(i) If there are more than two qualities, in a system of resale
markets, there is no set of N  1 vintage dependent prices
that supports the revealing strategy profile.
(ii) If there are only two qualities, then there exists an ex-post
efficient consumer equilibrium.
Simple Depreciation Model with Unobservable Quality
Consumers cannot observe the quality of a car without using it.
Theorem 1 :
(i) If there are more than two qualities, in a system of resale
markets, there is no set of N  1 vintage dependent prices
that supports the revealing strategy profile.
(ii) If there are only two qualities, then there exists an ex-post
efficient consumer equilibrium.
Remark: Inefficiency does not vanish in the limit as Δ → 0.
Simple Depreciation Model with Unobservable Quality
Consumers cannot observe the quality of a car without using it.
Theorem 1 :
(i) If there are more than two qualities, in a system of resale
markets, there is no set of N  1 vintage dependent prices
that supports the revealing strategy profile.
(ii) If there are only two qualities, then there exists an ex-post
efficient consumer equilibrium.
Remark: Inefficiency does not vanish in the limit as Δ → 0.
Proposition 1: If there exists K  1 resale markets and an equilibrium
strategy profile that yields the efficient allocation, then K  N and the
consumer equilibrium consists of the revealing strategy profile.
Simple Depreciation Model with Unobservable Quality
Consumers cannot observe the quality of a car without using it.
Theorem 1 :
(i) If there are more than two qualities, in a system of resale
markets, there is no set of N  1 vintage dependent prices
that supports the revealing strategy profile.
(ii) If there are only two qualities, then there exists an ex-post
efficient consumer equilibrium.
Remark: Inefficiency does not vanish in the limit as Δ → 0.
Proposition 1: If there exists K  1 resale markets and an equilibrium
strategy profile that yields the efficient allocation, then K  N and the
consumer equilibrium consists of the revealing strategy profile.
Proposition 1 and Theorem 1 if n ≥ 2 there is no efficient
consumer equilibrium under stochastic depreciation and resale.
Simple Depreciation Model with Unobservable Quality:
Why Resale is Inefficient?
Simple Depreciation Model with Unobservable Quality:
Why Resale is Inefficient?
For revealing strategies to be an equilibrium with resale markets,
type  ∗n must be just willing to be a vintage-n consumer ex-ante.
Simple Depreciation Model with Unobservable Quality:
Why Resale is Inefficient?
For revealing strategies to be an equilibrium with resale markets,
type  ∗n must be just willing to be a vintage-n consumer ex-ante.
Furthermore, he should be willing to sell the good as soon as it
depreciates to quality q n1 .
Simple Depreciation Model with Unobservable Quality:
Why Resale is Inefficient?
For revealing strategies to be an equilibrium with resale markets,
type  ∗n must be just willing to be a vintage-n consumer ex-ante.
Furthermore, he should be willing to sell the good as soon as it
depreciates to quality q n1 .
These two conditions imply that he should be willing to sell a
vintage-n good that just depreciated to quality q n1 and then buy
a vintage-n  1 good whose quality (in equilibrium) should be q n1 .
Simple Depreciation Model with Unobservable Quality:
Why Resale is Inefficient?
For revealing strategies to be an equilibrium with resale markets,
type  ∗n must be just willing to be a vintage-n consumer ex-ante.
Furthermore, he should be willing to sell the good as soon as it
depreciates to quality q n1 .
These two conditions imply that he should be willing to sell a
vintage-n good that just depreciated to quality q n1 and then buy
a vintage-n  1 good whose quality (in equilibrium) should be q n1 .
This cannot be optimal as by keeping the vintage-n car that is
of quality q n1 until it depreciates again, the consumer enjoys a
quality q n1 good which he can then sell for p n1.
Simple Depreciation Model with Unobservable Quality:
Why Resale is Inefficient?
For revealing strategies to be an equilibrium with resale markets,
type  ∗n must be just willing to be a vintage-n consumer ex-ante.
Furthermore, he should be willing to sell the good as soon as it
depreciates to quality q n1 .
These two conditions imply that he should be willing to sell a
vintage-n good that just depreciated to quality q n1 and then buy
a vintage-n  1 good whose quality (in equilibrium) should be q n1 .
This cannot be optimal as by keeping the vintage-n car that is
of quality q n1 until it depreciates again, the consumer enjoys a
quality q n1 good which he can then sell for p n1.
In contrast, if he buys a vintage n  1 car, he would still enjoy a
quality q n1 unit, but would only be able to sell it for p n2 .
Simple Depreciation Model with Unobservable Quality:
Efficiency of Rental
Simple Depreciation Model with Unobservable Quality:
Efficiency of Rental
Consider rental contracts that specify an instantaneous rental
price r n the consumer pays for renting vintage-n.
Simple Depreciation Model with Unobservable Quality:
Efficiency of Rental
Consider rental contracts that specify an instantaneous rental
price r n the consumer pays for renting vintage-n.
The consumer can keep renting the same unit as long as she
wishes, and stop paying the rental fee the moment she wishes
to return the unit (without any cancellation fees).
Simple Depreciation Model with Unobservable Quality:
Efficiency of Rental
Consider rental contracts that specify an instantaneous rental
price r n the consumer pays for renting vintage-n.
The consumer can keep renting the same unit as long as she
wishes, and stop paying the rental fee the moment she wishes
to return the unit (without any cancellation fees).
Consumers are not allowed to rent units they returned in the
past in order to prevent them from strategically returning cars.
Simple Depreciation Model with Unobservable Quality:
Efficiency of Rental
Consider rental contracts that specify an instantaneous rental
price r n the consumer pays for renting vintage-n.
The consumer can keep renting the same unit as long as she
wishes, and stop paying the rental fee the moment she wishes
to return the unit (without any cancellation fees).
Consumers are not allowed to rent units they returned in the
past in order to prevent them from strategically returning cars.
Theorem 2 Under rental, there is a consumer equilibrium under
asymmetric information that has the same allocation, strategies,
and instantaneous rental prices as under observable quality.
Simple Depreciation Model with Unobservable Quality:
Efficiency of Rental
Consider rental contracts that specify an instantaneous rental
price r n the consumer pays for renting vintage-n.
The consumer can keep renting the same unit as long as she
wishes, and stop paying the rental fee the moment she wishes
to return the unit (without any cancellation fees).
Consumers are not allowed to rent units they returned in the
past in order to prevent them from strategically returning cars.
Theorem 2 Under rental, there is a consumer equilibrium under
asymmetric information that has the same allocation, strategies,
and instantaneous rental prices as under observable quality.
The efficent rental contracts must have indeterminate duration.
Simple Depreciation Model with Unobservable Quality:
Adding Supply
Simple Depreciation Model with Unobservable Quality:
Adding Supply
There is a unit measure of producers, each of whom has an
opportunity to produce a single unit of the good at a cost c in
every period.
Simple Depreciation Model with Unobservable Quality:
Adding Supply
There is a unit measure of producers, each of whom has an
opportunity to produce a single unit of the good at a cost c in
every period.
Firms have the same instantaneous discount factor as consumers.
Simple Depreciation Model with Unobservable Quality:
Adding Supply
There is a unit measure of producers, each of whom has an
opportunity to produce a single unit of the good at a cost c in
every period.
Firms have the same instantaneous discount factor as consumers.
Let Ry the per-unit expected present value of revenue as a
function of the total industry output y.
Simple Depreciation Model with Unobservable Quality:
Adding Supply
There is a unit measure of producers, each of whom has an
opportunity to produce a single unit of the good at a cost c in
every period.
Firms have the same instantaneous discount factor as consumers.
Let Ry the per-unit expected present value of revenue as a
function of the total industry output y.
Let y ∗ be the zero profit (first-best) output level (Ry ∗  c).
Simple Depreciation Model with Unobservable Quality:
Adding Supply
There is a unit measure of producers, each of whom has an
opportunity to produce a single unit of the good at a cost c in
every period.
Firms have the same instantaneous discount factor as consumers.
Let Ry the per-unit expected present value of revenue as a
function of the total industry output y.
Let y ∗ be the zero profit (first-best) output level (Ry ∗  c).
A fraction y of firms produce each period. Active firms offer rental
contracts at instantaneous prices r n y Nn0 . The remaining 1 − y
firms are inactive.
Simple Depreciation Model with Unobservable Quality
and Endogenous Supply: Efficiency of Rental
Simple Depreciation Model with Unobservable Quality
and Endogenous Supply: Efficiency of Rental
Theorem 3: The following constitutes a market equilibrium
for any Δ  0:
Simple Depreciation Model with Unobservable Quality
and Endogenous Supply: Efficiency of Rental
Theorem 3: The following constitutes a market equilibrium
for any Δ  0:
(i) firms produce the first-best output y ∗ , and offer N  1
vintage dependent rental contracts at the instantaneous
rental prices determined as in the observed quality model,
Simple Depreciation Model with Unobservable Quality
and Endogenous Supply: Efficiency of Rental
Theorem 3: The following constitutes a market equilibrium
for any Δ  0:
(i) firms produce the first-best output y ∗ , and offer N  1
vintage dependent rental contracts at the instantaneous
rental prices determined as in the observed quality model,
(ii) for every n  0, . . . , N, consumer types  ∈  ∗n ,  ∗n−1  rent
vintage-n cars and only keep cars of quality q n , where the
cutoffs are determined as in the observed quality model.
Simple Depreciation Model with Unobservable Quality
and Endogenous Supply: Efficiency of Rental
Theorem 3: The following constitutes a market equilibrium
for any Δ  0:
(i) firms produce the first-best output y ∗ , and offer N  1
vintage dependent rental contracts at the instantaneous
rental prices determined as in the observed quality model,
(ii) for every n  0, . . . , N, consumer types  ∈  ∗n ,  ∗n−1  rent
vintage-n cars and only keep cars of quality q n , where the
cutoffs are determined as in the observed quality model.
Not only the efficient sorting is achieved in equilibrium, but
the first-best amount of output is supplied.
General Depreciation Model with Unobservable Quality
General Depreciation Model with Unobservable Quality
In the general depreciation model:
General Depreciation Model with Unobservable Quality
In the general depreciation model:
(1) initial quality is uncertain;
General Depreciation Model with Unobservable Quality
In the general depreciation model:
(1) initial quality is uncertain;
(2) depreciation can occur in more than one step.
General Depreciation Model with Unobservable Quality
In the general depreciation model:
(1) initial quality is uncertain;
(2) depreciation can occur in more than one step.
The key distinction from the simple one-step depreciation
model is the fact that sorting now requires experimentation.
General Depreciation Model with Unobservable Quality
In the general depreciation model:
(1) initial quality is uncertain;
(2) depreciation can occur in more than one step.
The key distinction from the simple one-step depreciation
model is the fact that sorting now requires experimentation.
For instance, highest-valuation consumers need to try
several units before finding one of quality q 0 .
General Depreciation Model with Unobservable Quality
In the general depreciation model:
(1) initial quality is uncertain;
(2) depreciation can occur in more than one step.
The key distinction from the simple one-step depreciation
model is the fact that sorting now requires experimentation.
For instance, highest-valuation consumers need to try
several units before finding one of quality q 0 .
It is impossible now to obtain the efficent allocation: the
first consumer of the good consumes the ‘wrong’ quality
with positive probability.
General Depreciation Model with Unobservable Quality
Approximate Efficiency
General Depreciation Model with Unobservable Quality
Approximate Efficiency
Theorem 4
(i) There exists Δ ∗  0, s.t., for all Δ ∈ 0, Δ ∗ , there is a
consumer equilibrium with rental prices r n  Nn0 where
consumer types  ∈  n ,  n−1  rent vintage-n cars and only
keep cars of quality q n .
General Depreciation Model with Unobservable Quality
Approximate Efficiency
Theorem 4
(i) There exists Δ ∗  0, s.t., for all Δ ∈ 0, Δ ∗ , there is a
consumer equilibrium with rental prices r n  Nn0 where
consumer types  ∈  n ,  n−1  rent vintage-n cars and only
keep cars of quality q n .
(ii) Furthermore, as Δ → 0, cutoff types and instantaneous
rental prices converge to their observable quality counterparts:
 n →  ∗n and r n → r ∗n for all n  0, . . . , N.
Discussion
Discussion
• Full information efficient allocation can be achieved in a
competitive equilibrium even under asymmetric
information.
Discussion
• Full information efficient allocation can be achieved in a
competitive equilibrium even under asymmetric
information.
• This result indicates that inefficiency in standard adverse
selection models of durables is not due solely to
asymmetric information, but to a combination of other
restrictions of trading possibilities.
Discussion
• Full information efficient allocation can be achieved in a
competitive equilibrium even under asymmetric
information.
• This result indicates that inefficiency in standard adverse
selection models of durables is not due solely to
asymmetric information, but to a combination of other
restrictions of trading possibilities.
• Challenges of Rental Implementation:
Discussion
• Full information efficient allocation can be achieved in a
competitive equilibrium even under asymmetric
information.
• This result indicates that inefficiency in standard adverse
selection models of durables is not due solely to
asymmetric information, but to a combination of other
restrictions of trading possibilities.
• Challenges of Rental Implementation:
- Agents could form a coalition to strategically return cars.
Discussion
• Full information efficient allocation can be achieved in a
competitive equilibrium even under asymmetric
information.
• This result indicates that inefficiency in standard adverse
selection models of durables is not due solely to
asymmetric information, but to a combination of other
restrictions of trading possibilities.
• Challenges of Rental Implementation:
- Agents could form a coalition to strategically return cars.
- Moral hazard is likely to be severe problem in rental.